bug#32460: 27.0.50; diff-mode sometimes doesn't refine hunks

bug#32460: 27.0.50; diff-mode sometimes doesn't refine hunks

[Juri Linkov]

[-- Attachment #1: Type: text/plain, Size: 729 bytes --]

It's difficult to debug a sporadic problem when automatic diff
refinement doesn't display refined overlays on some diff hunks
occasionally.  Looking at the code, it seems there is a possible
race condition when diff--font-lock-refined calls diff-hunk-next
that sets a timer that later might remove and try to re-add again
overlays already added by the direct call of diff--refine-hunk
from diff--font-lock-refined.  Also diff-hunk-next prevents
multiple consequent calls from refining a set of hunks
by checking diff--auto-refine-data.

To avoid such double refinement, this patch let-binds
diff-auto-refine-mode to nil before calling diff-hunk-next
for non-interactive navigation (it calls diff--refine-hunk
explicitly below):


[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: diff-refine.1.patch --]
[-- Type: text/x-diff, Size: 1253 bytes --]

diff --git a/lisp/vc/diff-mode.el b/lisp/vc/diff-mode.el
index b91a2ba45a..175687f184 100644
--- a/lisp/vc/diff-mode.el
+++ b/lisp/vc/diff-mode.el
@@ -2077,7 +2088,9 @@ diff--font-lock-refined
                   (point) 'diff--font-lock-refined nil max)))
     (let* ((min (point))
            (beg (or (ignore-errors (diff-beginning-of-hunk))
-                    (ignore-errors (diff-hunk-next) (point))
+                    (ignore-errors (let ((diff-auto-refine-mode nil))
+                                     (diff-hunk-next))
+                                   (point))
                     max)))
       (while (< beg max)
         (let ((end
@@ -2092,7 +2105,10 @@ diff--font-lock-refined
               (overlay-put ol 'modification-hooks
                            '(diff--font-lock-refine--refresh))))
           (goto-char (max beg end))
-          (setq beg (or (ignore-errors (diff-hunk-next) (point)) max)))))))
+          (setq beg (or (ignore-errors (let ((diff-auto-refine-mode nil))
+                                         (diff-hunk-next))
+                                       (point))
+                        max)))))))
 
 (defun diff--font-lock-refine--refresh (ol _after _beg _end &optional _len)
   (delete-overlay ol))

bug#32460: 27.0.50; diff-mode sometimes doesn't refine hunks

[Charles A. Roelli]

tags 32460 fixed
close 32460 27.1
quit

As Juri suggested, the fix for Bug#32991 probably fixed this bug.

bug#32460: 27.0.50; diff-mode sometimes doesn't refine hunks

[Stefan Monnier]

> As Juri suggested, the fix for Bug#32991 probably fixed this bug.

As part of the fix for Bug#32991, I rewrote the code that visits "all"
hunks in a region, because that code had a bug which basically made it
skip every other hunk: in practice with jit-lock (where the granularity
is small enough that the region considered rarely contains more than one
hunk), it tended to "arbitrarily" miss some hunks every once in a while
(regardless of diff-hunk-next).

Not sure if this error was the cause of bug#32460 since you mention
interaction with diff-hunk-next, but I think it's indeed better to close
the bug for now and reopen it (or a new one) if it re-appears.


        Stefan